Professional Documents
Culture Documents
Received 17 July 2017; received in revised form 12 October 2017; accepted 29 October 2017
Handling Editor: A. Giaccari
Available online 11 November 2017
KEYWORDS Abstract Background and aims: This study compared the accuracy of the FreeStyle Libre (Ab-
Type 1 diabetes; bott, Alameda, CA) and Dexcom G4 Platinum (DG4P, Dexcom, San Diego, CA) CGM sensors.
Continuous glucose Methods and results: Twenty-two adults with type 1 diabetes wore the two sensors simulta-
monitoring (CGM); neously for 2 weeks. Libre was used according to manufacturer-specified lifetime (MSL); DG4P
Flash glucose was used 7 days beyond MSL. At a clinical research center (CRC), subjects were randomized to
monitoring; receive the same breakfast with standard insulin bolus (standard) or a delayed and increased (de-
Accuracy; layed & increased) bolus to induce large glucose swings during weeks 1 and 2; venous glucose was
Hypoglycaemia; checked every 5e15 min for 6 h. Subjects performed 4 reference fingersticks/day at home. Ac-
Rate of change curacy was assessed by differences in mean absolute relative difference (%MARD) in glucose levels
compared with fingerstick test (home use) and YSI reference (CRC). During home-stay the Libre
MARD was 13.7 3.6% and the DG4P MARD 12.9 2.5% (difference not significant [NS]). With
both systems MARD increased during hypoglycaemia and decreased during hyperglycaemia,
without significant difference between sensors. In the euglycaemic range MARD was smaller with
DG4P [12.0 2.4% vs 14.0 3.6%, p Z 0.026]. MARD increased in both sensors following delayed
& increased vs. standard bolus (Libre: 14.9 5.5% vs. 10.9 4.1%, p Z 0.008; DG4P: 18.1 8.1% vs.
13.1 4.6%, p Z 0.026); between-sensor differences were not significant (p Z 0.062). Libre was
more accurate during moderate and rapid glucose changes.
Conclusions: DG4P and Libre performed similarly up to 7 days beyond DG4P MSL. Both sensors
performed less well during hypoglycaemia but Libre was more accurate during glucose swings.
Trial registration: The study was registered in ClinicalTrials.gov (NCT02734745) April 12, 2016.
ª 2017 Published by Elsevier B.V. on behalf of The Italian Society of Diabetology, the Italian Society
for the Study of Atherosclerosis, the Italian Society of Human Nutrition, and the Department of
Clinical Medicine and Surgery, Federico II University.
* Corresponding author. Department of Medicine, University Hospital of Padova, Via Giustiniani 2, 35128 Padova, Italy.
E-mail address: daniela.bruttomesso@unipd.it (D. Bruttomesso).
https://doi.org/10.1016/j.numecd.2017.10.023
0939-4753/ª 2017 Published by Elsevier B.V. on behalf of The Italian Society of Diabetology, the Italian Society for the Study of Atherosclerosis, the Italian Society of Human
Nutrition, and the Department of Clinical Medicine and Surgery, Federico II University.
FreeStyle Libre and Dexcom G4 Platinum sensors 181
psychological benefits in T1D subjects treated with interstitial glucose levels every minute. By scanning the
continuous subcutaneous insulin infusion (CSII) and mul- reader over the sensor one obtains a real-time glycaemic
tiple daily insulin injection (MDI) [4e12]. Most glucose value, the glycaemic trend over the preceding 8 h, the
sensors, however, require calibration, have short life time, direction towards which glucose levels are moving and
measure glucose in the interstitial fluid and cannot always glucose rate of change. The sensor lasts 14 days.
replace blood glucose self-monitoring (SMBG). The Dexcom G4P consists of a small sensor inserted
The FreeStyle Libre Flash glucose monitoring system under the skin which measures subcutaneous glucose
(FGM, Abbott Diabetes Care, Alameda, CA) is factory- concentration every 5 min and sends readings to a
calibrated and has 14 days of wear lifetime but lacks receiver. Twice-daily calibration with fingerstick blood
real-time alarms when limit glucose values are exceeded. glucose is required. The manufacturer specified lifetime
It is intended as a replacement for the SMBG, except dur- (MSL) for the sensor is 7 days, however it is possible to
ing times of rapid glucose changes, when hypoglycaemia extend sensor use for a further week by reactivating it [15].
requires confirmation or when displayed glucose values
are not in accordance with symptoms. Procedures
A study evaluating the accuracy of Libre system at home
for 10e14 days [13] and another [14], comparing the ac- Patients were trained on the use of the two devices at the
curacy of Libre to that of two commercially available sen- CRC. Sensors were simultaneously placed on the back of
sors, found that Libre had comparable accuracy. To date, no one upper arm (Libre) and on the abdominal region (DG4P).
study has compared Libre’s performance with that of
comparators during rapid glucose swings or beyond sensor Home phase
certified wear lifetime. During at-home CGM use, subjects were instructed to
The most commonly used CGM sensor in Italy is Dex- perform SMBG at 4 times per day (before meals and at
com G4 Platinum (DG4P, Dexcom Inc., San Diego, CA), bed-time) using the blood glucose (BG) meter built into
which is provided by the health care system as adjunctive the hand-held Libre reader. Patients were instructed to
to SMBG. The Libre system is purchased by patients or scan the sensor immediately after each BG test and to
provided by the health care system without adjunctive based their insulin dosage adjustments on the BG reading.
SMBG for clinical decision-making. Patients calibrated the DG4P according to manufactures’
The aim of this study was to compare the accuracy of specification against the BG measurements. After 7 days,
Libre vs. DG4P in adults with T1D as assessed by mean subjects began a second 7-day session with the original
absolute difference (% MARD) during glycaemic excursions DG4P sensor to assess accuracy beyond the manufacturer
induced in at a clinical research center and during home specified lifetime. In case of sensor failure, loss of signal,
use. skin issues or accidental dislodgment, the sensor was
substituted. After 14 days, data from FreeStyle Libre and
Methods DG4P were downloaded.
when glucose was <3.0 mmol/L (<54 mg/dL) or at physi- used to compare normally and not normally distributed
cian’s discretion and sampling was every 5 min in case of variables. The means of more than two group were
hypoglycaemia. Patients left the CRC at 2:00 p.m. compared with one-way analysis of variance (ANOVA). The
level of significance accepted was 0.05 using two-tailed
Sensor accuracy assessments tests. Statistical analysis was performed using MATLAB,
Statistics Toolbox (Release 2016a, The MathWorks, Inc.,
The systems were tested over 14 days of continuous use at- Natick, Massachusetts, United States).
home and during two six-hour CRC admissions scheduled
at day 3e5 and 9e11 (Fig. S1). Accuracy was expressed as Results
mean absolute relative difference (%MARD). Plasma
glucose values were used as reference during the CRC Subject disposition
visits and SMBG values were used as reference during the
home phase. Sensor data were matched with venous (CRC Twenty-four adults with type 1 diabetes were enrolled.
assessments) or capillary (home use assessment) BG One subject was excluded from analysis as home data
measurements. When CGM and reference values could not could not be uploaded; one subject left the study for poor
be obtained at the same time, CGM values were linearly device acceptance. Twenty-two subjects completed the
interpolated to match reference values, provided CGM study: 13 female/9 male (10 CSII/12 MDI), mean age
values were obtained within 5 min of reference values. 36.3 12.9 years (mean SD), diabetes duration
MARD analyses were performed separately for break- 18.9 11.1 years, BMI 23.5 2.7 kg/m2, HbA1c 7.3 0.75%
fast with normal bolus vs. breakfast with increased and (56.7 8.19 mmol/mol).
delayed bolus, and for week 1 vs. week 2 of sensor use. Ten subjects had breakfast with delayed and increased
MARD was calculated over the entire glycaemic range and bolus during the first week of sensor use, 12 during the
for glucose values: <3.9 mmol/L (70 mg/dL), and 3.9e10 second week. The Libre device was scanned 14,4 6 time/
mmol/L (70e180 mg/dL) and >10 mmol/L (>180 mg/dL). day.
Sensor accuracy was also assessed by calculating both the
percentage of data points in zones A and A þ B of the Home phase
Clarke Error Grid (CEG) [16] and the proportion of values
that fulfilled ISO 15197:2013 criteria (percentage of sensor As shown in Fig. 1, there was no significant difference in
data within 15% of reference value for glucose concen- overall accuracy between Libre and DG4P during at-home
trations 5.6 mmol/L (100 mg/dL), and within 0.8 mmol/ use. Both systems demonstrated similar performances in
L (15 mg/dL)) of reference value for glucose concentrations the hypoglycaemic and hyperglycaemic ranges; however,
<5.6 mmol/l (100 mg/dL). DG4P showed better accuracy in the euglycaemic range.
During CRC sessions, MARD was also calculated by Both systems showed worse accuracy in the hypo-
consolidating BG references into five groups based on the glycaemic range.
rate of change (ROC), calculated as the first-order differ- Comparison of week 1 to week 2 showed similar ac-
ence between the current and the previous sample divided curacy with both sensors during home use (Fig. 1). The
by the time distance between the two. The five ROC ranges MARD for both systems worsened in week 2, but not
were: increasing glucose, >0.08 mmol/L/min (>1.5 mg/dL/ significantly (Libre: 13.0 3.7% week 1 vs. 14.6 4.9%
min) and >0.03 to 0.08 mmol/L/min (>0.5 to 1.5 mg/dL/ week 2, p Z 0.080; DG4P: 12.5 4.1% week 1 vs.
min); stable glucose, 0.03 to 0.03 mmol/L/min (0.5 to 13.4 3.2% week 2, p Z 0.092) (Table S1).
0.5 mg/dL/min); and decreasing glucose, >0.03 to As shown in Table S1, Libre’s accuracy worsened
0.08 mmol/L/min (>0.5 to 1.5 mg/dL/min), and >0.08 significantly during days 11e14 compared with days 1e10
mmol/L/min (>1.5 mg/dL/min). Sensor accuracy based on (12.6 6.0% on days 1e10 vs 15.0 8.2% on days 11e14,
ROC values was measured considering all CRC sessions p Z 0.006). DG4P had the worst accuracy on days 1 and 8
(with and without induced glycaemic excursions), and (14.5 7.1% vs. 11.7 7.4%, p Z 0.015), which was ex-
separating first from second week visits. pected; CGM performance on day 1 of sensor wear is
Because sensor accuracy was expected to be lower on known to be worse than performance on subsequent days.
day 1 of sensor insertion [15,17] and during the second
week of monitoring (due to possible sensor performance CRC studies
degradation), MARD analyses were also performed
comparing: day 1 vs. all the other days; days 1 and 8 vs. all Induced glycaemic excursions
other days; week 1 vs. week 2; and combined days 1e10 MARD increased significantly in both the Libre and DG4P
vs. combined days 11e14. Data are presented as systems following delayed and increased bolus at break-
mean standard deviation. fast; the betweenesensor difference was not significant
(Table 1). Libre showed a lower MARD vs. DG4P during
Statistical analysis hyperglycaemia, whether induced by a breakfast with
delayed and increased bolus or following a breakfast with
Univariate analyses were performed for descriptive sta- correctly administered insulin bolus (Table 1). Overall
tistics. The t-test and the Wilcoxon signed-rank test were MARD values (with/without induced glycaemic
FreeStyle Libre and Dexcom G4 Platinum sensors 183
Figure 1 Top: Mean absolute relative difference (MARD) per day 95% confidence interval for Libre or DG4P. Bottom: Mean absolute relative
difference (MARD) between Libre or DG4P glucose readings and capillary glucose reference concentration at home.
Table 1 Mean absolute relative difference (%MARD) between Libre and DG4 P glucose readings and reference glucose concentration in venous
blood in subjects with type 1 diabetes receiving a breakfast with standard or delayed and increased insulin bolus.
Glucose profile Breakfast w/standard insulin bolus Breakfast w/delayed and increased insulin bolus
Libre DG4P Data pairs P value Libre DG4P Data pairs P value
Overall 10.9 4.1 13.1 4.6 424 0.055 14.9 5.5 18.1 8.1 710 0.062
Hypoglycaemia 10.8 6.9 12.9 11.9 9 0.824 21.7 14.4 27.3 22.9 87 0.148
<3.9 mmol/L (<70 mg/dL)
Euglycaemia 13.3 4.9 13.7 6.7 245 0.894 17.3 6.9 18.5 7.2 362 0.421
3.9e10 mmol/L (70e180 mg/dL)
Hyperglycaemia 7.8 4.5 11.2 5.1 170 0.010 10.2 4.9 14.5 6.1 261 0.031
>10 mmol/L (>180 mg/dL)
184 F. Boscari et al.
excursions) were similar during week 1; however, Libre and 1 sensor was interrupted at day 12 for acetaminophen
MARD was significantly lower during week 2 (Table S2). assumption by the subject.
Libre MARD values were also significantly lower in the
hyperglycaemic range during weeks 2 (Table S2). Safety and adverse events
No sensor failure requiring removal or infection at the
Clark error grid analysis and fulfilment of ISO insertion site were reported.
15197:2013 criteria
There was no difference in the systems’ clinical perfor- Discussion
mance, most values fell within the clinically acceptable
error zones (A þ B) (Fig. 2, Table S3). This study compared the Libre and DG4P continuous
A notable proportion of glucose values from Libre and glucose monitoring systems during at-home use and dur-
DG4P fulfilled ISO 15197:2013 accuracy criteria: 70.2% and ing induced glycaemic excursions. It is important to note
73.5%, respectively, during home phase; 74.1% (Libre) and that the systems compared are not strictly equivalent,
65.0% (DG4P) during CRC assessment with normal insulin since Libre data can be used in place of SMBG for insulin
bolus and by 69.5% (Libre) and 57.0% (DG4P) during CRC adjustments, while DG4P is considered adjunctive to
assessment with increased and delayed insulin bolus SMBG. Furthermore, MSL is 14 days for Libre and only 7
(Table S3). days for DG4P. Despite these differences, we felt it was
important to compare the two systems because they are
Rates of glucose change widely used in Italy, and because many patients wear
Analysis of accuracy during glucose swings was possible DG4P beyond MSL.
only where YSI values were available. No significant dif- In designing the study protocol, we relied on evidence
ference in sensor accuracy was observed when glucose showing that DG4P accuracy remains unchanged 7 days
was stable 0.03 to 0.03 mmol/L (0.5 to 0.5 mg/dL/min), beyond MSL [15]. As shown in our study, we found that the
while Libre was significantly more accurate during periods Libre and Dexcom G4 sensors functioned with similar ac-
of moderate and rapid glucose changes (Table 2). Accuracy curacy during home use for 2 weeks. Surprisingly, DG4P
among both sensors within the DG4P MSL (7 days), was accuracy was even better in the euglycaemic range, which
similar except during glucose changes at >0.08 mmol/L/ is likely due to the 2 daily calibrations recommended by
min (>1.5 mg/dL/min) (Table 2). the manufacturer.
We also found that Libre’s accuracy decreased between
Sensors lifetime days 11 and 14, which is an important consideration for
Mean sensor lifetime was 13.45 days (1.3) for DG4P and patients who base insulin dosing on sensor readings only.
13.5 (1.2) for Libre. Eighteen (86.4%) Libre sensors func- The increase in MARD and variability of DG4P on day 8 was
tioned for the full 14 days; 2 sensors were removed at day due to sensor recalibration, which is intrinsic to the Dex-
10 and 1 at day 13 for adhesive issues and 1 sensor was com algorithm [15].
replaced at day 3 due to accidental dislodgment. Sixteen With both systems, no difference in accuracy was found
(63.3%) DG4P functioned for the full 2 weeks, 1 sensor was comparing week 1 vs. week 2 (in CRC and at home). As
removed at day 9 and 4 at day 13 for adhesive problems, expected, the accuracy of both systems worsened
Figure 2 Clarke Error Grid Analysis. Left: Libre (grey dots) and DG4P (black dots) vs. capillary measurements during 14 days at home. Right: Libre
(grey dots) and DG4P (black dots) vs. venous measurements after delayed and increased insulin administration.
FreeStyle Libre and Dexcom G4 Platinum sensors 185
P value
<0.001
<0.001
time delay between plasma and interstitial glucose [18].
0.019
0.039
0.248
Additionally, both sensors deviated approximately 20%
from reference value in the hypoglycaemic range during
Table 2 Mean absolute relative difference (%MARD) between Libre or DG4P readings and reference venous blood glucose according to different rates of change of glucose concentration. home use, indicating that low glucose readings should
Data pairs
routinely be confirmed by SMBG.
In studies done at the CRC during the first week Libre
144
157
71
55
66
and DG4P had similar accuracy across all glucose values
but Libre had a lower MARD when glucose decreased by
25.5 12.2
14.6 12.1
14.6 10.9
14.9 12.7
13.9 18.7
>0.08 mmol/L/min (>1.5 mg/dL/min). During the second
MARD (%)-YSI sessions
12.4 13.6
8e14 days
10.6 9.0
12.6 8.5
10.8 9.0
leagues, comparing Libre with DG4P and Medtronic Enlite
over 12 h in an inpatient setting under real-life conditions
Libre
0.362
0.702
0.684
0.049
180
106
99
72
13.0 15.3
13.2 14.5
18.3 19.4
13.5 12.6
12.7 10.6
13.7 13.6
<0.001
<0.001
<0.001
0.010
0.083
321
337
172
14.0 13.7
14.3 13.6
16.9 18.9
MARD (%)-All YSI sessions
12.9 11.0
13.2 13.3
1e14 days
11.8 9.9
Conclusion
>0.08 (>1.5)
Decreasing
Increasing